Abstract: Based on the FP_growth association rules, an improved a lgorithm with the dynamically weighted characteristic was put forward. Accord ing to their important degree, the items within the transaction database were divided into 5 grades, AHP was used to construct the judgmental matrix, and the eigenvector was calculated. At the same time, in order to create the FP_tree and to measure its important degree, taking the vector we got as weight, a synthetic consideration was made between the weight and the number items ap peared in the transaction database. Finally the frequent item sets and associati on rules were found. As the process of getting weight can be dynamically changed by domain expert, not only can the more meaningful rules be mined, but also a lot of lower weighted item sets were got rid of during the early time of algorithm, so the efficiency of our algorithm was improved.

Key words: data mining, association rule, FP_growth, weighted tree, AHP